CVPR 2025 Tutorial: Multimodal Mathematical Reasoning

Overview

Welcome to the CVPR 2025 Tutorial on Multimodal Mathematical Reasoning! Artificial intelligence is progressively advancing toward systems capable of addressing complex reasoning tasks across multiple modalities. At the intersection of computer vision, natural language processing, and mathematical understanding, multimodal mathematical reasoning has emerged as a pivotal area of research. This tutorial presents recent advancements that empower AI models to interpret diagrams, comprehend mathematical problems expressed in natural language, and reason over structured visual and textual inputs within a unified framework. We will provide a comprehensive overview of current progress in areas including benchmark development, multimodal alignment, pretraining and fine-tuning strategies for mathematical tasks, and reinforcement learning with GRPO-based optimization. In addition, we examine techniques such as inference-time scaling and tool-augmented reasoning, which further extend the capabilities of contemporary systems. This tutorial aims to elucidate the fundamental challenges in this domain and inspire future research toward developing interpretable and general-purpose multimodal reasoning systems.

Date: June 12, 2025
Time: 1:15 PM – 5:00 PM CDT (Nashville local time)
Location: Nashville, TN
In-person attendance:Room 202C
Virtual attendance: Click here
Convert to your timezone: Click here

Program (CDT)

Time (CDT)	Speaker	Topic
1:15 – 1:30	Xiangliang Zhang & Arman Cohan	Motivation & Challenges of Multimodal Math Reasoning
1:30 – 2:10	Pan Lu	Benchmarking Multimodal Math Reasoning
2:10 – 3:20	Yilun Zhao & Tianyu Yang	Foundation Models and Alignment for Multimodal Math Reasoning
3:20 – 4:00	Minhao Cheng	Reinforcement Learning & GRPO
4:00 – 4:40	Weijia Shi	Inference-Time Scaling & Tool-Augmented Reasoning
—	—	Q&A & Ending